Modern Geometry: Fall, 2021: Part 1 – The Basics

October 3, 2021

Introduction
These notes represent the first part of the lecture notes for the course. They
are divided into four sections. The first starts with a brief review of the
history of geometry and then introduces the basic objects that will concern
us for the entire course: smooth and complex manifolds, their structure
sheaves of functions and Lie Groups. The second section introduces vector
bundles with the prime example of the tangent bundle and other tensor
budnles of a smooth manifold. We introduce vector fields with their Lie
bracket and prove the existence and uniqueness of integral curves for a vector
feild..

1 Basic Definitions of Sheaves, Manifolds, and Lie

groups
1.1 Classical Geometry: Euclidean geometries
Geometry began with Euclid’s axioms for plane geometry. This is a synthetic
geometry. The basic notions are points and lines: with the usual Euclidean
axioms: through any two points there is a unique line, two lines meet in at
most one point, and if they do not meet then they are call parallel; given a
line and a point not on the line there is a unique parallel to the line through
the point, etc.
The standard realization of this geometry as the usual Cartesian plane
with coordinates (x, y) and distance function
d((x, y), (x0 , y 0 )) = (x − x0 )2 + (y − y 0 )2 .
p

Points are the usual points of the space and lines are straight lines infinite
in both directions. These are geodesics, in the sense that they are length
minimizing curves. This model satisfies all the Euclidean axioms

1
1.1.1 Early Generalizations
Dropping the parallel axiom there is spherical geometry: the model space is
the unit sphere in Cartesian 3-space; the geodesics are great circles. All lines
meet: on the sphere, hence there are no parallel lines. Unfortunately lines
meet in two points rather than the one required by the Euclidean axiom,
but this defect is remedied by passing to the projective plan where antipodal
points are identified. Again the great circles are geodesics, i.e., locally length
minimizing curves.
Much later in the 19th came Lobachevskian geometry where through
a point not on a line there are infinitely many lines parallel to the given
line. A model for this geometry is the open unit disk in the plane with
lines being circular arcs orthogonal to the unit circle. These lines are length
minimizing for the Poincaré metric on the open unit disk, or equivalently
the (open) upper-half space model where lines are vertical lines and circular
arcs perpendicular to the x-axis.
All of these geometries are homogeneous in the sense that given any two
points there is an isomorphism of the geometry taking the first point to the
second, and given two lines through a point there is an isomorphism of the
geometry fixing the point and sending the first line to the second.

1.1.2 What Is Modern Geometry?

The word geometry comes from the roots geo meaning earth and metric
meaning measure. It was originally conceived of as the study of measure-
ment on the earth (distances, areas, etc). In modern geometry the objets
consist of a space (almost alway a topological space) together with some
extra structure that allows for some type of measurement on the space. The
most direct generalization of Euclidean geometry is Riemannian geometry
where the spaces are smooth manifolds and notion of measurement is a local
(infinitessimal) one of angels and lengths. The study also goes under the
name of differential geometry, especially when considering curves, surfaces
and other submanifolds of a given ambient manifold and their curvatures
inside the ambient manifold. More exotic is Spin geometry which arises
from the fact that the orthogonal group has a non-trivial double covering,
giving rise to spinor fields . Related to Riemannian geometry is conformal
geometry where the local measurement is angles only.
There is symplectic geometry where the measurement is of areas of sur-
faces and higher dimensional volumes of even dimensional submanifolds.
There is the closely related study of contact geometry, which concerns odd

2
dimensional manifolds (‘boundaries’ of symplectic manifolds).
In the context of spaces endowed with complex structure there is an ana-
logue of Riemannian geometry called hermitian geometry, and an important
subcase of Kähler geometry. More special than this are Calabi-Yau geom-
etry (meaning Kähler and complex sympletic) and Hyperkähler geometry
based on the quaterions.
Other modern uses of the term geometry include complex geometry
where the spaces are endowed with a complex structure, and algebraic ge-
ometry where the spaces are defined by polynomial functions (and those
polynomial functions are part of the structure of the space). In these uses of
the word geometry, there is extra structure but it is not directly a measure-
ment as we usual conceive of it, rather geometry in this context refers to the
fact that there are interesting pictures of the spaces in question associated
with the extra structure.
Most of the spaces that we will encounter will be smooth manifolds, but
singularities (and singularity development) are often important aspects of
geometry. This is especially true in algebraic geometry. We will only touch
on singularities in this course.

1.2 Topological Manifolds

A topological manifolds is a space (without singularities) on which various
types of geometry exist as extra structure.
In every dimension n ≥ 0 there is a Euclidean space, denoted Rn . Its
points are ordered n-tuples x = (x1 , . . . , xn ) of real numbers. There is the
Euclidean distance function
v
u n
uX
d(x, y) = t (xi − yi )2 .
i=1

This is metric in the usual topological sense, e.g., it is a continuous func-

tion of two variables, symmetric in the two variables; it satisfies the triangle
inequality; and if d(x, y) = 0, then x = y. In the usual way, this metric de-
fines a topology. The resulting topological space endowed with this distance
function is Euclidean n-space.

Definition 1.1. An n-dimensional manifold is a paracompact, Hausdorff

space with the property that every point has a neighborhood homeomor-
phic to an open subset of Rn . An n-dimensional manifold with boundary
is a paracompact Hausdorff space with the property that every point has

3
a neighborhood homeomorphic either to an open subset of Rn or of the
half-space x1 ≥ 0 in Rn .

Invariance of domain implies that each point of a compact manifold with

boundary with the second type of neighborhood in which it maps to a point
of the boundary {x1 = 0} does not have a neighborhood of the first type.
The subset of points satisfying this condition form a closed subset and inherit
the structure on an (n − 1)-dimension manifold without boundary.
Alternatively, we could define a manifold by giving a Hausdorff topolog-
ical space M and a countable atlas of open subsets {Ui ⊂ M } in M (atlas
in the sense that ∪i Ui = M ), together with homeomorphisms ϕi : Ui → Vi
where the Vi are open subsets in Rn or in {x1 ≥ 0}.
Or we could start with open subsets Vi ⊂ Rn or in {x1 ≥ 0} ⊂ Rn and
for each ordered pair {i, j} an open subset Wi,j ⊂ Vi and homeomorphisms
hi,j : Wi,j → Wj,i (called transition functions or overlap functions) that sat-
isfy hj,k ◦ hi,j = hi,k on h−1 i,j (Wj,k ) ∩ Wi,k for all triples {i, j, k}. Also, we
assume that for all i we have Wi,i = Vi and hi,i = IdVi and for all {i, j} we
have hj,i = h−1 i,j . In this presentation we have to assume that the quotient
space
∪i Vi /{x ∈ Wi,j ≡ hi,j (x) ∈ Wj,i }
is a Hausdorff space.
Using the first definition, for each open subset Ui ⊂ M , the homeo-
morphism ϕi : Ui → Vi ⊂ Rn or ⊂ {x1 ≥ 0} produces local coordinates
{x1 , . . . , xn } on Ui . These are the pull back via ϕi of the usual Cartesian
coordinates restricted to the open Vi in Rn . These local coordinates deter-
mine and are determined by the hopmeomorphism of U to an open subset
V in Rn or {x1 ≥ 0}.
Of course, these are only local coordinates. We define Wi,j ⊂ Vi as
ϕi (Ui ∩ Uj ) and the overlap function hi,j : Wi,j → Wj,i to be ϕj ◦ ϕ−1
i . These
functions give the transition formulae for one set of local coordinates in
terms of the other on the overlap.

1.3 Sheaves
We begin with a very important general notion: that of a sheaf on a topo-
logical space X. Let Op(X) be the category with objects the open subsets
of X and the morphisms the inclusions of open sets. A presheaf (of real
vector spaces) is a contravariant functor from Op(X) to the category of
vector spaces. In down-to-earth terms this means an assignment for each
open subset U ⊂ X of a vector space F(U ) and if V ⊂ U a ’restriction

4
map’ rU,V : F(U ) → F(V ) compatible with compositions in the sense that
rV,W ◦ rU,V = rU,W .
A presheaf F is a sheaf if it satisfies the following gluding axiom: For
every open cover U = ∪i Ui of an open set U ⊂ X the following sequence is
exact: Q
i6=j (ri,j −rj,i )
Q
i ri
Y Y
0 → F(U ) −→ (Ui ) −→ F(Ui ∩ Uj ),
F i6=j

where ri = rU,Ui and ri,j = rUi ,Ui ∩Uj . Notice that by the composition axiom
the composition of the two mappingsQis zero. Hence, the sheaf condition
Q that any element in the kernel of i,j (ri,j − rj,i ) is in the image under
is
Qa unique element of F(U ). This means that given an element
i ri of
{si } ∈ i F(Ui ) there is an element s ∈ U whose restriction to Ui is equal
to si for each i if and only if the {si } satisfy the compatibility condition that
si |Ui ∩Uj = sj |Ui ∩Uj for all i, j, and furthermore such an s is unique.

1.4 Smooth manifolds

Definition 1.2. We say that a real-valued function ψ : V → R on an open
subset of a Euclidean space is smooth if it has continuous partial derivatives
of all orders. Similarly, a function on an open subset of {x1 ≥ 0} is smooth
if it has a smooth extension to some open subset of Rn . More generally,
a map between open subsets of Euclidean space or Euclidean half-space
ψ : V → W is smooth iffy the composition of ψ with all the Euclidean
coordinate functions restricted to W are smooth.

Notice that the composition of smooth functions is smooth and that a

function is smooth if and only if it is smooth in some neighborhood of every
point of the domain.

Definition 1.3. Let M be an n-dimensional manifold. A smooth atlas

for an n-dimensional manifold M consists of an atlas {Ui ⊂ M } and of
homeomorphisms ϕ : Ui → Vi to open subsets of Vi ⊂ Rn or V ⊂ {x1 ≥ 0}
with the property that the overlap functions hi,j : Wi,j → Wj,i are smooth.
This is equivalent to requiring that the differentials of the hi,j are C ∞ at
every point of their domains. It is equivalent to say that the local coordinates
for one coordinate are smooth functions of the coordinates from any other
coordinate patch on their overlap.
Given a smooth atlas we say that a function f : U → R from an open
subset of M to R is smooth (with respect to the given smooth atlas) if for
each p ∈ U and any Ui in the atlas containing p the function f ◦ϕ−1i is smooth

5
function on ϕi (Ui ∩ U ) ⊂ Vi ⊂ Rn . Notice this condition is independent of
the choice of Ui in the smooth atlas containing p. Let C ∞ (U ) denote the
vector space of smooth functions on U . Clearly if U 0 ⊂ U then restriction
of functions defines a linear map C ∞ (U ) → C ∞ (U 0 ) and if U 00 ⊂ U 0 the
restriction from U to U 00 is the composition of the restriction from U to U 0
followed by the restriction from U 0 to U 00 . Hence, this defines a presheaf
on M . Since a function is determined by its values and a function is C ∞ if
and only if it is C ∞ in a neighborhood over every point of its domain, this
presheaf satisfies the gluing axiom and hence is a sheaf. It is the sheaf of
smooth functions defined by the smooth atlas. Two smooth atlases define
the same smooth structure on M if their sheaves of smooth functions are
identical. Hence, a smooth structure on M is an equivalence class of smooth
atlases.
It is easy to see that two smooth atlases define the same smooth structure
iff the coordinate functions of all the charts in one atlas are smooth functions
with respect to the other atlas and vice-versa.
Definition 1.4. An n-dimensional smooth manifold is a topological n-
dimensional manifold endowed with a smooth structure. A smooth manifold
is a disjoint union of n-dimensional smooth manifolds for various n.
It is an elementary exercise to show a sheaf on an n-dimensional manifold
M is the sheaf of smooth functions with respect to some smooth atlas on
M if and only if it is locally isomorphic to the sheaf of smooth functions on
Rn .
Definition 1.5. If M and N are smooth manifolds, then a continuous map
ϕ : M → N is smooth iff for every open set U ⊂ N and ever f ∈ C ∞ (U ) we
have ϕ∗ f ∈ C ∞ (ϕ−1 (U )).
Smooth manifolds and smooth maps between them form a category, the
smooth category; an isomorphism in this category is called a diffeomorphism.
This category has (countable) sums given by disjoint union and finite
products given by Cartesian product.
A smooth submanifold of a manifold M is a subset X ⊂ M with the
property that for every x ∈ X there is a coordinate chart U ⊂ M with local
coordinates (x1 , . . . , xn ) with the property X ∩ U is given by the vanishing a
a subset of these coordinates. Clearly, a submanifold inherits the structure
of a smooth manifold such that the inclusion is a smooth map.
The usual product of topological spaces produces a product on the cate-
gory of smooth manifolds: the smooth coordinate charts for M × N are the

6
product of the charts of M with those for N . This extends to an associative
product of finitely many smooth manifolds.

1.5 Complex manifolds

A complex atlas on a manifold M is an atlas of coordinate charts ϕi : Ui →
Vi where the Vi are open subsets of Cn (identified with R2n in the usual
way) with the property that the overlap functions hi,j : Wi,j → Wj,i are
holomoprphic, meaning that the differential of the hi,j at every point of its
domain is a complex linear map from Cn to itself. (Complex manifolds do
not have boundaries in the sense that smooth ones do, since the boundary is
real codimension 1 whereas all complex manifolds have even real dimension.)
Completely analogously to the smooth case, a complex atlas defines a
sheaf of complex-valued holomorphic functions on open subset of M . Here,
the vector spaces assigned to open subsets are complex vector spaces, in-
stead of real vector spaces. A complex structure is an equivalence class of
complex atlases, where two complex atlases are equivalent if and only if they
define the same sheaf of holomorphic functions. Complex manifolds form a
category with the morphisms being continuous maps that pull local holo-
morphic functions on the range back to local holomorphic functions on the
range.
There is a natural product (finite products) in the category of complex
manifolds which on the level of topological spaces is the usual product.
Notice that underlying a complex manifold of (complex) dimension n is
a smooth manifold of (real) dimension 2n.

1.6 Lie Groups

A Lie group is a smooth manifold G that has a group structure given by
a multiplication map µ : G × G → G and inverse map ι : G → G with the
property that both these structure maps are smooth.
Examples of Lie Groups include countable discrete groups, and in par-
rticular finite groups, the circle, the general linear group, denoted GL(n, R)
of all linear isomorphisms of Rn , the orthogonal group, the group of trans-
lations of a real vector space. The 3-sphere has a group structure because it
is identified with the group of quaterions of unit norm and these are closed
under quarerionic multipliciation.
A morphism between Lie groups is a c smooth map that is also a group
homomorphism. A Lie subgroup of a Lie group G is a smooth submanifold

7
that is also a subgroup. The orthogponal group O(n) is a Lie subgroup of
GL(n, R).
The group GL(n, C) of complex linear isomorphisms of Cn is also a Lie
group. The unitary group U (n) consisting of all complex linear transfor-
mations that preserve the standard hermitian inner product on Cn is a Lie
subgroup of GL(n, C).
A finite product of Lie groups is a Lie group.

Definition 1.6. Let G be a Lie group and let M be a smooth manifold.

Then a smooth action of G on M is an action

µ: G × M → M

that is a smooth map.

Exercise: Show that if G is a finite group acting freely on a smooth maniold

M there is a unique smooth structure on the quotient M/G with the prop-
erty that the smooth functions on M/G are exactly those whose pull back
to M are smooth functions. The smooth functions on M/G are identified
in this way with the smooth functions on M invariant under G.
The antipodal action of {±1} on Rn is defined by µ(±, (x1 , . . . , xn )) =
(±x1 , . . . , ±xn ). It is a smooth action. whose restriction to the unit sphere
S n−1 is a free smooth action.
The quotient of S n by the antipodal action is denoted RP n and is called
the real projective n-space. It is naturally identified with the space of real
lines (1-dimensional linear suspaces) of Rn+1 . It follows that |rP n is a
smooth manifold and its smooth functions are identified with the smooth
functions on Sn invariant under the antipodal action.

Definition 1.7. A complex Lie group G is a complex manifold with a group

structure with the property that the multiplication map µ : G × G → G and
the inverse map ι : G → G are holomorphic mappings (i.e., morphisms of
complex manifolds)..

Any closed complex submanifold of G(n, C) that is closed under matrix

multiplication is a complex Lie group. These are called linear algebraic
groups (defined over the complex numbers). The matrix coordinates are
holomorphic functions on any linear algebraic group. It follows easily from
the fact holomorphic functions on compact complex manifolds are locally
constant that any positive dimensional linear algebraic group (over C) is
non-compact.

8
 An example of a compact complex Lie group is given by C/Λ where Λ is
a lattice in C, i.e., a subgroup isomorphic to Z⊕Z generated by two elements
of C that are linearly independent over R. By a complex-linear change of
coordinate on C we can assume that the subgroup is generated by 1 and an
element of the form τ = a + bi with b > 0. The resulting element τ in the
upper half-plane is well-defined up to fraction linear transformations
ατ + β
τ 7→
γτ + δ
where  
α β
γ δ
is an element of SL(2, Z), the group of integral 2 × 2 matrices with deter-
minant 1.

1.7 Examples
Suppose that F : Rn+1 → R is a smooth function and at each x ∈ F −1 (0)
the differential dfx : R→ R is non-zero (meaning that for each x ∈ F −1 (0)
there is at least one partial derivative of F at x that is non-zero. Then by
the implicit function theorem, for each x ∈ F −1 (0) here is a neighborhood
U (x) ⊂ F −1 (0) and one of the coordinates, say for simplicity of notation
xn+1 such that Ux is the graph of a function xn+1 = ϕx (x1 , . . . , xn ) defined
on an open subset Vx of the coordinate hyperplane defined by xn+1 = 0.
This produces a local coordinate chart of the neighborhood Ux . The union
of all such charts is easily seen to define a smooth atlas. The sheaf of C ∞ -
functions determined by this atlas is exactly the restriction to F −1 (0) of the
sheaf of C ∞ -functions on Rn+1 .
Indeed the smooth function F does not need to be defined on all of Rn+1 ,
just on an open subset of Rn+1 .
Taking X
f (x1 , . . . , xn+1 ) = 1 − x2i
i

we see that the unit sphere Sn ⊂ Rn+11 is a smooth n-dimensional manifold.

More generally, if U ⊂ R n+k is an open subset and F : U → Rk is
a smooth function with the property that at each point x ∈ F −1 (0) the
differential DFx : Rn+k → Rk has rank k (ie., is a surjective linear map), the
F −1 (0) inherits the structure of a smooth n-manifold.
Let O(n) be the group of real orthogonal matrices. Then O(n) is a
smooth manifold of dimension n(n − 1)/2. The reason is that O(n) ⊂ M (n),

9
 2
where M (n) is the space of n × n matrices which is identified with Rn .
The orthogonal group is defined by the n(n + 1)/2 conditions that say the
columns of the matrices have inner product +1 with themselves and the
inner product between two different columns is zero. The gradients of these
n(n + 1)/2 defining equations are linearly independent at any element of
O(n) ⊂ M (n × n). This establishes that O(n) is a smooth submanifold of
M (n × n) and hence of GL(n, R). Hence, it is a Lie group.
We have embeddings Rn−1 ⊂ Rn and these lead to embddings O(n−1) ⊂
O(n) making O(n−1) a sub-Lie group of O(n); it is the subgroup of matrices
that fix en = (0, 0, . . . , 0, 1) ∈ Rn . Thus the map O(n) → S n−1 that sends
A ∈ O(n) to Aen is a smooth action and it identifies O(n)/O(n − 1) with
S n−1 . Similarly. O(n)/O(n − k) is identified with the space of orthonormal
k-frames {f1 , · · · , fk } in Rn . That is to say hfi , fj i = δi,j . These spaces
inherit manifold structures from the smooth structure on O(n). As function
on the quotient space is smooth if and only if its pullback to O(n) is smooth.
There is a similar story for the unitary groups U (n) ⊂ GL(n, C) of
matrices that preserve the standard hermitian form on Cn . The quotient of
U (n)/U (n − 1) is identified with the unit sphere in Cn ; i.e., S 2n−1 .
Notice that even though the unitary group is defined via complex struc-
tures it is not a complex Lie group. Indeed the odd unitary groups are of
odd dimension.
Real projective n-space. Consider the space RP n of linear subspaces
of dimension 1 (lines) in Rn+1 . Each such line meets unit sphere S n in a
pair of antipodal points. Thus, the space is identified with the quotient
of S n by the antipodal action of the group of two elements. The smooth
structure is the induced one. A point x ∈ RP n is determined any any point
(x0 , . . . , xn ), different from 0, in the line. Any two such points differ by
coordinate-wise multiplication by a non-zero real number. We denote the
”projective” coordinates by [x0 , . . . , xn ], not all zero, where implicitly we
can replace the coordinates by any non-zero multiple. This is a smooth
n-dimensional manifold.
Complex projective n-space. CP n is the space of 1-dimensional
complex linear subspaces of Cn+1 . The complex projective coordinates are
[z0 , . . . , zn ] ,not all zero; two-such are equivalent if they differ by coordinate-
wise multiplication b a non-zero complex numbers. A complex atlas is given
by Ui , 0 ≤ i ≤ n consists of all points whose projective coordinates have
non-zero entry in position i.
The homeomorphism ϕi : Ui → Cn defined by

ϕi ([z0 , . . . , zn ]) = (z0 /zi , . . . . , zi−1 /zi .zi+1 /zi , . . . , zn /zi ).

10
One sees easily that the overlap functions are holomorphic.
Suppose that f (z0 , . . . , zn ) is a homogeneous polynomial of degree d with
the property that ∇f is non-zero at every point (z0 , . . . , zn ) 6= (0, . . . , 0)
satisfying f (z0 , . . . , zn ) = 0, then the solutions of f = 0 is a union of complex
lines and hence defines a subset of CP n . An easy computation shows that
the locus in CP n is a comnplex codimension-1 submanifold of CP n , called a
smooth hypersurface of degree d.. A hypersurface of degree 1 is isomorphic
to CP n−1 .

11
 .

2 Vector Bundles, the tangent bundle and other

tensor bundles of smooth manifolds, vector fields
2.1 Vector Bundles
2.2 The basic definition
In this lecture all real vector spaces are implicitly assumed to be finite di-
mensional. There are analogues for infinite dimensional spaces but one has
to specify topologies or other structures, e.g., Banach spaces, Hilbert spaces,
or Frechet spaces.

Definition 2.1. Let X be a topological space. A family of real vector spaces

over X is a continuous map π : E → X and two maps, a sum map

µ : E ×X E → E

and a scalar multiplication map

R×E →E

both commuting with the natural projections to X making each fiber π −1 (x)
a real vactor space. A real vector space is the same thing as a family of real
vector spaces over a point.

Families of real vector spaces over topological spaces are the objects of
a category. The morphisms are commutative diagrams

E −−−−→ E 0
 
πy π0 y
 

X −−−−→ X 0
commuting with the structure maps. There is also a pull back mapping: if
π : E → X is a family of real vector spaces and f : Y → X is a continuous
map, then there is a family f ∗ E → Y and a morphism
f ∗ E −−−−→ E
 
 
y y
Y −−−−→ X

12
defined by f ∗ E = E ×X Y with the naturally induced projection and structure
maps. In particular, given a family of real vector spaces over X and a
subspace Y ⊂ X, there is the restriction of the family to Y .
We say that a family of real vector spaces is locally trivial if for every
x ∈ X the restriction of the family to a neighborhood U of x is isomorphic
to the trivial family V × U where V is a real vector space and the structure
maps are pulled back from the usual structure maps on V . Said another
way E|U is isomorphic to the pullback of a family over a point.
Definition 2.2. Areal vector bundle is a locally trivial family of real vector
spaces.
There is an analogue of coordinate charts for real vector bundles. If
π : E → X is a real vector bundle, then there is an open covering {Uα }α
of X, vector spaces Vα , and isomorphisms ϕα : E|Uα → Vα × Uα of vector
bundles.
The overlap functions are then isomorphisms (Uα ∩ Uβ ) × Vα → (Uα ∩
Uβ ) × Vβ covering the identity of (Uα ∩ Uβ ). Such an isomorphism is given
by a continuous map hαβ : Uα ∩ Uβ → Iso(Vα , Vβ ). If X is connected, then
we can choose identifications of all the Vα with a fixed vector space V . Then
the overlap functions are given by continuous maps hαβ : Uα ∩ Uβ → GL(V ).
These satisfy the cocycle conditions:
• hαα = IdUα
• hβα = h−1
αβ

• hβγ ◦ hαβ = hαγ on Uα ∩ Uβ ∩ Uγ .

There is full subcategory of real vector bundles in families of vector
spaces.
A subbundle of a vector bundle E → X is a family of subspaces Fx ⊂ Ex
for all x ∈ X, such that there are local trivializations of E that also give
local trivializations of F.

2.3 Examples
Associated to any real vector space V for every topological space X there
is the trivial vector bundle V × X pulled back from the bundle V → {∗}
by the unique map X → {∗}. As another example in trivial vector bundle
S 1 × R2 → S 1 consider the sublocus of all (θ, x) where x lies on the line
R · (cos(θ/2), sin(θ/2)). This is a subvector bundle whose total space is
isomorphic to the open Möbius band.

13
2.3.1 Sheaves of Sections
Definition 2.3. A sheaf of rings over a topological space X is a contravari-
ant functor from the category of open subsets of X to rings satisfying he
gluing axiom.
For example, the sheaf of real valued functions on X, denoted O(X) is
a sheaf of rings over X.
Definition 2.4. Given a sheaf of rings R over X a sheaf of R-modules is a
sheaf of abelian groups on X such that for each open set U ⊂ X the group
associated to U has the structure of an R(U )-module and the restriction
maps are compatible with the ring and module structures.
Definition 2.5. Asection of a vector bundle π : E → X over an open set
U is a map s : U → E satisfying π ◦ s = IdU . These from an abelian group
under the structure maps of the vector bundle (called fiberwise addition and
scalar multiplication).
Proposition 2.6. Let E → X be a vector bundle. The functor that assigns
to each open set U ⊂ X the abelian group of sections of this bundle over
U is a sheaf of abelian groups. Fiberwise scalar multiplication makes this
sheaf into a sheaf of O(X)-modules. [Recall that O(X) is the sheaf of rings
of local real-valued functions on X.]
Remark 2.7. A vector bundle is isomorphic to a trivial bundle X × V if
and only if its sheaf of sections is a free module of finite rank.

2.4 Smooth Vector Bundles

Let M be a smooth manifold. A smooth vector bundle over M is a vector
bundle E → M together with a smooth structure on E such that the projec-
tion to M is a smooth submersion (surjective differential) with smooth local
trivializations such that the structure maps for addition and scalar multipli-
cation are smooth maps. In this case by O(M ) we mean the sheaf of smooth
functions on M . It is a sheaf of rings and a subsheaf of the sheaf of continu-
ous functions. A smooth section of a smooth vector bvundle π : E → M over
an open set U is a smooth function s : U → E with π ◦ s = IdU . These form
a sheaf of abelian groups under fiberwise addition and scalar multipliciation
and indeed form a module over the sheaf of smooth functions. A smooth
vector bundle is trivial as a smooth vector bundle if and only if this module
is a a free module of finite rank.
The overlap functions for smooth local trivializations of a smooth vector
bundle are smooth maps from hij : Ui ∩ Uj → GL(n, R).

14
2.5 Complex Vector Bundles
We can replace real vector spaces by complex vector spaces and produce
the categories of a family of complex vector spaces over a topological space
and of complex vector bundles over topological spaces. There is a forgetful
functor that associates to a family of complex vector spaces the underlying
family of real vector spaces and associates to a complex vector bundle the
underlying real vector bundle. In the case of complex vector bundles the
overlap functions for local trivializations are maps from the intersections of
the open sets in the base to GL(n, C).
As above when we work over a smooth manifold we have the category
of smooth complex vector bundles. The sheaf of smooth sections is a sheaf
of modules over the sheaf of smooth complex-valued functions on the base.
When we work over a complex manifold we have the category of so-called
holomorphic vector bundles where the total space is a complex manifold, the
projection mapping is a holomorphic surjection, the local trivializations are
holomorphic isomorphisms and the structure maps are holomorphic. In this
case the overlap functions for holomorphic trivializations are holomorphic
maps hij : Ui ∩ Uj → GL(n, C). The sheaf of local holomorphic sections is a
sheaf of modules over the sheaf of holomorphic functions on the base and the
vector bundle is holomorphically trivial. if and only if this sheaf of modules
is free of finite rank.

2.6 Tangent Bundle

Let M be a smooth n-manifold. Denote by Ix (M ) the ideal of the ring of
smooth functions on M that vanish at x.
Lemma 2.8. Ix (M )/(Ix (M ))2 is an n-dimensional real vector space.
Proof. For any open subset U ⊂ M containing x, the restriction map on
smooth functions induces a ring homomorphism sending Ix (M ) to Ix (U ).
The kernel consists of functions on M vanishing near x. The map is not
onto, but any C ∞ -function on U is the sum of the restriction of a C ∞ -
function on M and a function on U vanishing near x. [Use a bump function
to damp a given function on U to 0 outside a compact set and extend by 0
to all the rest of M .]
Functions on M and U vanishing a neighborhood of x are contained in
(Ix (M ))2 and (Ix (U ))2 , respectively. [Multiply by a function that vanishes
near x and is 1 off the support of the function.] It follows that the induced
linear map
Ix (M )/(Ix (M ))2 → Ix (U )/(Ix (U ))2

15
induced by restriction is an isomorphism.
Thus, it suffices to prove the result for any neighborhood of x. We choose
a coordinate chart containing x; i.e., a smooth isomorphism ϕ : U → V where
V is an open subset of Rn and x ∈ U . We can assume that ϕ(x) = 0. Cleary,
ϕ identifies Ix (U )/(Ix (U ))2 with I0 (V )/(I0 (V ))2 and the latter is identified
with I0 (Rn )/(I0 (Rn ))2 .
Let F be a smooth function on Rn vanishing at the origin. Then by the
chain rule
Z 1 X Z 1 ∂F (tx)
F (x) = ∇F (tx) · xdt = xi dt.
0 0 ∂xi
i
R1
We set hi (x) = 0 ∂F∂x(tx) ∂F (0)
P
i dt., so that F (x) = i xi hi (x) and hi (0) = ∂xi .
n 2
Thus, F ∈ (I0 (R )) if and only if hi (0) = 0 for all 1 ≤ i ≤ n. Hence, the
map F → (h1 (0), . . . , hn (0)) determines an isomorphism from I0 (Rn )/(I0 (Rn ))2
to Rn .

∂F
Notice that a basis for Tx (Rn ) are the linear maps F 7→ |
∂xi x
for 1 ≤
i ≤ n.

Definition 2.9. The tangent space to M at x ∈ M , denoted Tx M is the

vector space of linear maps from Ix (M )(/Ix M )2 to R. Notice that a basis
∂
for Tx (Rn ) is ( ∂x i )x for 1 ≤ i ≤ n.

Consider T M the union over all x ∈ M of the tangent spaces Tx (M ). We

give this family of vector spaces a topology so that it forms a smooth vector
bundle. First, if M is an open subset of Rn we define a global trivialization of
T M by using the basis ( ∂x∂ 1 , . . . , ∂x∂n ) at each point of M . This determines a
function ∪x∈M Tx M → M × Rn that is a linear isomorphism on each tangent
space, and hence endows T M with the structure of a smooth vector bundle.
More generally, for any manifold M this construction produces a smooth
vector bundle structure on ∪x∈U Tx (M ) for any coordinate chart U ⊂ M .
To see that this determines a global smooth bundle structure we must show
that the smooth trivializations on the tangent bundle over the intersection of
two charts determined by each of the charts differ by a bundle isomorphism
given by a smooth map from the intersection to GL(n, R). This follows
immediately from the fact that the overlap function is Uα ∩ Uβ → GL(n, R)
is given by (x, v) 7→ (hα,β (x), Dhα,β (x)(v), which is a smooth isomorphism.
We denote this smooth vector bundle T M , and call it the tangent bundle
of M .

16
 The tangent bundle construction is a functor from the category of smooth
manifolds and smooth maps to the category of smooth vector bundles and
smooth vector bundle maps. It assigns to a smooth manifold M its tangent
bundle T M and to a smooth map f : M → N the map that sends v ∈ Tx M
to Dfx (v). As before the chain rule gives a formula in local coordinates on
M near x and local coordinates on N near y for Df in a neighborhood of
x. This formula shows that Df is a smooth bundle map.
Definition 2.10. Let M be a smooth manifold. A vector field on M is
a smooth section of T M . In local coordinates (x1 , . . . , xn ) on a coordinate
∂
chart for M any vector field is given as i ui (x1 , . . . , xn ) ∂x
P
i for some smooth
functions ui .
Remark 2.11. Often, to simplify notation, we denote ∂/∂xi by ∂i .
Notice that if f : M → N is a diffeomorphism then the smooth bundle
map Df : T M → T N sends smooth local sections of T M to smooth local
sections of T N . Thus, there is a map induced by Df from the vector fields
on M to those on N . In particular, a self-diffeomorhism of M acts on the
vector fields on M .
Claim 2.12. Given a diffeomorphsims f : M → N with f (x) = y, the pull-
back map f ∗ on smooth functions defines a map Iy (N ) → Ix (M ). Using
local coordinates xi for M near x and y j for N near y and the resulting ba-
sis {∂/∂xi }(x) and {∂/∂y j }(y) for the dual spaces Tx M and Ty N the map
dual to
f ∗ : Iy (N )/(Iy (N ))2 → Ix (M )/)Ix (M ))2
is given by matrix multiplication by the matrix of partial derivatives of f at
x:  1 
∂y ∂y 1
1 (x) · · · ∂x n (x)
 ∂x
 · ··· · 

Dfx =  · ··· · 
 
 · ··· · 
 
∂y n ∂y n
∂x1
(x) · · · ∂x n (x)

Proof. Let ϕ ∈ Iy (N ). We write ϕ = j y ϕj where ϕj (y) = (∂ϕ/∂y j )(y).

P j
Then X
f ∗ϕ = y j (x1 , . . . , z n )ϕ(y(x1 , . . . , xn )).
j

Module (Ix (M ))2 we have

X
y j (x1 , . . . , xn ) = (∂y j /∂xi )(x)
i

17
so that in IX (M )/(Ix (M ))2 we see that
XX 
f ∗ϕ = xi (∂y j /∂xi )(x)(∂ϕ/∂y j )(y) .


j i

It follows that under the dual map

X
(∂/∂xi )(x) 7→ (∂y j )/∂xi )(x)(∂ϕ/∂y j )(y).
j

Hence, the matrix for the dual map in the given bases is the matrix of partial
derivatives for Df (x).

Proposition 2.13. Let f : M → N be a diffeomorphism; let χ be a vector

field on M and ϕ a smooth function on N . Then

Df (χ)(f ∗ ϕ)(f (x)) = χ(f ∗ ϕ)(x).

Said another way

Df (χ) = (f −1 )∗ (χ ◦ f ∗ ).

Proof. We have just established that the map Tx M → Ty N is given by the

usual differential Df (x). Of course, χ(f ∗ ϕ)(x) = χ(ϕ ◦ f )(x) which by the
chain rule is Df (x)(χ(ϕ))(f (x)). This equation for all x ∈ M is eqivalent
to the equation
f ∗ (Df (x)(χ(ϕ))) = χ(f ∗ ϕ).
Taking (f −1 )∗ of this equation gives the second equation in the propositioin.

This is the naturaity equation for the action of vector fields on smooth
functions under diffeomorphisms.
There is a generalization of this. Suppose that f : M → N is a smooth.
This means for each x ∈ M , there is a local coordinate system on an open
set U of N centered at f (x) such that f : f −1 (f (N )∩U ) → U is an embefding
with image the intersection of U with a coordinate hyperplane, In this case
there is an induced map Df : T N → T M covering embeds each Tx N as
the coordinate linear subspace of Tf (x) M using the trivialization of T M |U
coming from th given coordinates on U .

18
2.6.1 Lie Bracket of Vector Fields
A basic structure of this action is that vector fields form a Lie algebra defined
by
[χ1 , χ2 ](f ) = χ1 (χ2 (f )) − χ2 (χ1 (f )).
Lemma 2.14. As given by the above formula [χ1 , χ2 ] is a vector field.
P
Proof.PIt suffices to compute in local coordinates where χ1 = ai ∂i and
χ2 = j bj ∂j . Then
X
[χ1 , χ2 ](f ) = [ai ∂i (bj ∂j (f )) − bj ∂j (ai ∂i (f ))]
i,j
X
= [ai (∂i (bj ))∂j (f ) + ai bj ∂i (∂j (f )) − bj (∂j (ai ))∂i (f )) − bj ai ∂i (∂j (f ))].
i,j

Since cross partials are equal, the second derivative terms cancel out and we
are left with
X
[χ1 , χ2 ](f ) = ai (∂i (bj ))∂j (f ) − bj (∂j (ai ))∂i (f )
i,j
X
= {ai ∂i (bj )∂j − bj ∂j (ai )∂i }(f ).
i,j

This proves that the space of vector fields is closed under this bracket
operation. Since composition of vector fields as operators on smooth func-
tions is associative, it follows that the bracket operation makes the space of
vector fields into a Lie algebra; i.e., a vector space with a bilinear operation
satisfying

[χ1 , χ2 ] + [χ2 , χ1 ] = 0
[[χ1 , χ2 ], χ3 ] + [[χ3 , χ1 ], χ2 ] + [[χ2 , χ3 ], χ1 ] = 0.

Definition 2.15. The bracket of vector fields is usually called the Lie
bracket of vector fields. The second equation is called the Jacobi identity.
Since the action of vector fields on smooth functions is natural under
diffeomorphisms, meaning

Df (χ1 )(Df (χ2 ))(ϕ) = (f −1 )∗ χ1 (f ∗ Dχ2 )(ϕ)

= (f −1 )∗ χ1 (f ∗ (f −1 )∗ χ2 (f ∗ ϕ)) = (f −1 )∗ (χ1 (χ2 (f ∗ ϕ))).

19
It follows that the bracket operation is natural under diffeomorphisms. If
f : M → N is a diffeomorphism, then the differential of f , Df : T M → T N
is an isomorphism of smooth vector bundles. As such it sends vector fields
on M to vector fields on N preserving the bracket operation.

2.6.2 Computing Lie brackets along submanifolds

Suppose that f : N → M is a smooth embedding and let A, B be vector fields
on N . Pushing forward via f we get partial vector fields Df (A), Df (B)
defined along f (N ) ⊂ M . That is to say that at each point of x ∈ f (N ) we
have tangent vectors Df (A), Df (B) ∈ T xM . Indeed these tangent vectors
are tangent to the submanifold f (N ).

Lemma 2.16. Let A, e B

e be any vectors fields on M extending the partial
vector fields Df (A) and Df (B) respetively. Then at any point x ∈ f (N ) we
have
Df ([A, B])(f −1 (x)) = [A,
e B]
e x.

Proof. If suffices to work locally. We choose coordinate charts U for N near

x and U × U 0 for M near f (x) such that f is the natural identification
U → U × {0} ⊂ U × U 0 . In these coordinates the lemma is immediate.

2.7 The Lie Algebra of a Lie Group

Let G be a Lie group. Denote by g = Te G its tangent space at the identity.
A vector field χ on G is said to be left-invariant if for every g ∈ G the
differential of left multiplication by g leaves χ invariant.

Theorem 2.17. The left-invariant vector fields on G form a (real) vector

space, invariant under the bracket of vector fields, and hence a sub-Lie alge-
bra of the Lie algebra of all vector fields on G. A left-invariant vector field
is determined by its value at e ∈ G. Assigning to each left-invariant vector
field its value at the identity of G determines a (real) linear isomorphism be-
tween the vector space of left-invariant vector fields and g. Transporting the
Lie bracket structure via this isomorphism gives us a Lie algebra structure
on g.

Proof. The map g 7→ D(g·) defines a smooth action of G as bundle iso-

morphism of T G covering the let multiplication of G on itself. Thus, fixing
χe ∈ g, defining χg = D(g·)(χe ) determines a smooth vector field on G (the
orbit of χe under the given smooth action of G.) Clearly, it is left-invariant

20
and the only left-invariant vector field with agrees with χe ∈ Te G. This
establishes the isomorphism between left-invariant vector fields and g.
Since the given action of G on T G is by smooth vector bundle isomor-
phisms, it preserves the Lie bracket. If follows that the Lie bracket of two
left-invariant vector fields is itself left-invariant.

Definition 2.18. g with the bracket given by the above theorem is called
the Lie algebra of G. Typically the symbol g us used to refer to this Lie
algebra.

Examples. 1. Since GL(n, R) is an open subset of the vector space M (n×n)

of n × n matrices, the tangent space to GL(n.R) at any point is identified
withM (n×n). Let A ∈ M (n×n). For any g ∈ GL(n, R) we have the matrix
gA. This is a smoothly varying family of matrices and hence a vector field
over GL(n, R). It is the left invariant vector field whose value at Te GL(n, R)
is A.
2. Let G ⊂ GL(n, R) be a sub-Lie group. Then for any g ∈ G the linear
space Tg G is identified with a subspace of M (n × N ). For each g ∈ G and
for any A ∈ Te G ⊂ M (n × n) the matrix gA is contained in Te G and this
family defines the left invariant vector field on G whose value at Te G is A.
3. Let A ∈ M (n × n) be a non-zeo matrix. For any t ∈ R the exponential
series etA = exp(tA) is well-defined and converges. For every t ∈ R , the
determinant of etA equals etdet(A) and hence is non-zero. The map R →
GL(n, R) defined by by t 7→ etA is a homomorphism from R → GL(n, R)
whose differential at the identity sends (∂/∂t)|0 to A ∈ Te GL(n, R). This
is an integral curve for the left-invariant vector field on GL(n, R) whose
value at the origin is A. It follows that if A ∈ Te G for some Lie subgroup
G ⊂ GL(n, R), then this homomorphism takes values in G.

2.7.1 The Lie Algebra of GL(n, R) and its subgroups

The tangent space to GL(n, R) at the identity (and indeed at any point)
is M (n × n). We have seen that the left invariant vector fields are of the
form χ(g) = g · A for g varying over GL(n, R) and A ∈ M (n × n) fixed. We
wish to compute the Lie bracket on M (n × n) coming from the Lie bracket
of left-invariant vector fields. To do this we must compute [gH, gK](e) for
arbitrary H, K ∈ M (n×n).PRecall that if in local coordinate (x1 , . . . , xn ) we
have vector fields H(x) = i hi ∂i and K(x) = i k i ∂i then for any point p
P

21
in the coordinate patch we have
X X
[H, K](p) = H(k j )(p)∂j − K(hi )(p)∂i .
j i

In our case GL(n, R) is an open subset of M (n × n) which is a vector

space with coordinates xi,j .
Let i,j : M (n×n) → R be function that associates to a matrix its (i, j)th
coordinate For H ∈ M (n × n) viewed as a tangent vector at the identity
element we have two basic formula:
1. The derivation H is equal to i,j i,j (H)∂i,j |e .
P

d tH
2. H(f (g)) = dt |t=0 (f (e )).

The general formula for the bracket of gH = i,j (gH)i,j ∂i,j and gK =
P
P i,j
i,j  (gK)∂i,j gives
X X
[gH, gK](e) = H(i,j (gK))∂i,j |e − K(i,j (gH))∂i,j |e
i,j i,j
X d X d
= |t=0 i,j (etH K)∂i,j |e − |t=0 i,j (etK H)∂i,j |e
dt dt
i,j i,j
X X X
= i,j (HK)∂i,j |e − i,j (KH)∂i,j |e = i,j (HK − KH)∂i,j |e .
i,j i,j i,j

The last term is the derivation at e ∈ G given by (HK−KH) ∈ M (n×n).

This proves that the restriction of bracket of left-invariant vector fields on
GL(n, R) induces the Lie algebra structure on M (n × n) given by

[H, K] = HK − KH.

Corollary 2.19. If G ⊂ GL(n, R) is a sub Lie group, then Te G ⊂ M (n × n)

is a sub Lie algebra under the bracket A ⊗ B 7→ [A, B].

2.8 Tensors and Differential Forms

Associated to a (finite dimensional) real vector space V is the dual space,
denote V ∗ consisting of all linear functions on V . Given a finite dimensional
vector space V there is the graded algebra, the tensor algebra on V , denoted
T (V ). It is ⊕n≥0 T n (V ) where

T n (V ) = V
| ⊗ ·{z
· · ⊗ V} .
n times

22
Juxtaposition of tensors defines an associative algebra structure. There is
also the exterior algebra, denoted Λ∗ (V ), which is the quotient of the tensor
algebra by the two-sided ideal generated by

v1 ⊗ v2 + v2 ⊗ v1 = 0.

If V has dimension n, then Λ∗ (V ) is of dimension 2n . If {e1 , . . . , en } is a

basis for V then {ei1 ∧ · · · ∧ eik } for all 1 ≤ i1 < i2 < · · · < ik ≤ n with
k ≥ 0 form a vasis for Λ∗ (V ).
All these operations extend from vector spaces to vector bundles, smooth
vector bundles, and holomorphic vector bundles. Dual to the tangent bundle
T M of a smooth manifold is the cotangent bundle, denoted T ∗ M . Sections
of this bundle are called differential 1-forms on M . In local coordinates
(x 1 n subset U ⊂ M a differential 1-form is written is
P , . . . ,1x ) onnan openi i i
i fi (x , . . . , x )dx where {dx }1≤i≤n is the dual basis to ∂/∂x . In a dif-
1 n
ferent set of coordinates (y , . . . , y ) the transformation law for the formula
for a differential 1-form is determined by
X ∂y j
dy j = dxi .
∂xi
i

More generally, a differential k-form is a section of Λk T ∗ M . The sum

over all k of the differential k-forms make a graded algebra with over the
ring a smooth functions multiplication given by wedge product. This graded
algebra is denoted Ω∗ (M ). Indeed the differential forms form a sheaf of
graded algebras over the sheaf of rings of smooth functions. An element of
Ωk (M ) is a differential form of degree k. (In particular Ω0 (M ) is the ring
of smooth functions.) A differential k-form is determined by its values on
k-tuples of vector fields. Furthermore, given a function on k-tuples of vector
fields it is evaluation of a differential k-form if and only if the function
is skew-symmetric under interchange of any pair of vector fields and the
function is linear over multiplication by smooth functions in each variable.
There is an important extra piece of structure that comes from dualizing
the Lie bracket on vector fields. The general principal is that the dual to
the.Lie bracket is a differential on the exterior algebra of the dual. In this
context we define d : Ω0 (M ) → Ω1 (M ) by

df (χ) = χ(f ).

In local coordinates df = i ∂i (f )dxi . Then we define d : Ω1 (M ) → Ω2 (M )

P
by
dω(χ1 , χ2 ) = χ1 (ω(χ2 )) − χ2 (ω(χ1 )) − ω([χ1 , χ2 ]).

23
Direct computation shows that dω is a skew symmetric operation bilinear
over the action of smooth functions on each factor. This means that dω is
a 2-form.
Since any differential form can be written as a sum of terms of the
form f dg1 ∧ dg2 ∧ · · · ∧ dgk for smooth functions f, g1 , . . . , gk , there is a
unique extension of d on functions and one-firms to an operation Ωk (M ) →
Ωk+1 (M ), for all k, satisfying linearity over R and the Leibnitz rule for
homogeneous differential forms:
d(α ∧ β) = dα ∧ β + (−1)|α| α ∧ dβ,
where |α| is the degree of α.
Dual to the Jacobi identity, is the identity that d2 (f ) = 0. This equation
can be established by direct computation in local coordinates and again uses
the equality of cross partial derivatives. Using the Leibintz rule one shows
that for any differential form α we have d2 (α) = 0. Again this can be
established by direct computation in local coordinates.
All of this structure is summarized by saying that Ω∗ (M ) is a differential
graded algebra: i.e., a graded algebra with a linear operator d raising degree
by 1, satisfying d2 = 0 and the Leibnitz rule.
If f : M → N is a smooth map, then there is an induced map of smooth
vector bundles Df : T M → T N . Suppose that ω is a differential 1-form
on N , i.e., a section of the cotangent bundle T ∗ N . It defines a smooth
bundle map ω : T N → N × R. The composition Df ◦ ω is then a smooth
bundle map T M → N × R covering the map f : M → N . The fiber product
of this map with the projection T M → M defines a smooth map T M →
M ×N (N × R) = M × R. This is a differential form on M , usually denoted
f ∗ ω.
In local coordinate (x1 , . . . , xn ) on M and y 1 , . . . y k ) on N , if ω = j µj ∗
P

y 1 . . . , y k )dy i , then
!
X X ∂y j
f ∗ω = µj (y 1 (x), . . . , y k (x)) (f (x)dxi .
∂xi
j i

More generally, there is a unique extension of this pullback of differential

1-forms, to a pullback of all differential forms compatible with wedge prod-
uct. It turns out (see the problem set) that pullback of differential forms
commutes with d.

2.9 Flows
A vector field on M generates, at least locally, a flow on M .

24
Definition 2.20. Let γ : (a, b) → M be a smooth map of an open interval
to M . Then Dγt0 (∂/∂t)) ∈ Tγ(t)0 M . These tangent vectors form a tangent
field defined along the curve. Given a vector field χ on M we say that γ is
an integral curve for χ if Dγt0 (∂/∂t) = χ(γ(t0 )) for all t0 ∈ (a, b).

Proposition 2.21. Given a vector field χ on M and a point x0 ∈ M there

is an  > 0 and an integral curve γ : (−, ) → M for χ with γ(0) = x0 . Any
two such agree on the intersection of their domains of definition. Further-
more, given a smooth map ρ : X → M there is an open neighborhood W of
X × {0} in X × R such that the intersection of W with each line {x} × R
is an open interval and such that for every x ∈ X the integral curve with
initial condition ρ(x) exists on the interval W ∩ {x} × R and these integral
curves define a smooth map ρ̂ : W → M .

Proof. In local coordinates χ is expressed as i fi (x1 , . . . , xn )(∂/∂xi ). The

P
equation that a curve γ : (−, ) → M is required to satisfy is γ̇ i (t) =
f i (γ(t)). The usual theorem on local existence and uniqueness of solutions
to vector-valued ODE’s and smooth variation with parameters gives the
result.

Now let us apply this to the family of initial conditions given by the
identity map M → M . We conclude that there is a neighborhood W of
M × {0} in M × R and a map Φ : W → M such that (i) the restriction
of Φ to M × {0} ⊂ W is the identity map of M to itself and (ii) the
restriction of Φ to the intersection of W with {x} × R is an integral curve
for the vector field χ. Now let us suppose that M is compact. In this case
there is  > 0 such that M × (−, ) ⊂ W . This allows us to assume that
W = M × (−, ). Thus, for each t ∈ (−, ) the map Φi = Φ|M ×{t} is a map
M → M with Φ0 = IdM . Since Φt varies smoothly with t, possibly after
making  > 0 smaller, we can assume that the differential of Φt is everywhere
an isomorphism; i.e., Φt is a local diffeomorphism, for all t ∈ (−, ). On the
other hand by the uniqueness of flow lines, it follows that Φt is one-to-one.
That is to say for every t ∈ (−, ), the map Φt is a diffeomorphism and flow
lines starting at any point of M extend in both directions for at least time
.
This means that in fact the flow lines are defined for all time and they
define a flow Φ : M × R → M , the flow generated by ch in the sense that
∂Φ
(x, t) = χ(Φ(x, t)).
∂t
One sees easily that the Φt define a group homomorphism from the additive

25
group R to the group of diffeomorphisms of M with the C ∞ -topology; i.e.,
Φs ◦ Φt = Φs+t and the Φt vary smoothly with t.
This shows that a vector field on a compact manifold M can be integrated
to a homomorphism of R to the group of diffeomorphisms with the C ∞ -
topology. Such homomorphisms are called flows.

26
3 A: Distributions and the Frobenius Theorem
Definition 3.1. A k-dimensional distribution on a smooth manifold M is
a smoothly varying family of k-planes in T M .

Equivalently, we could require that for each point x ∈ M the distribution

is given by the span of k smooth sections of the tangent bundle, sections
that are linearly independent at every point.

Definition 3.2. A k-dimensional foliation in a smooth manifold M is

an atlas of coordinates, called flow boxes, of the form U ⊂ Rk × Rn−k
with coordinates (x, y) such that the overlap functions hi,j are of the form
(gi,j (x, y), ki,j (y)). The level sets of the y-coordinate are called the local
leaves of the flow box. As we pass from one flow box to another the local
leaves match up.

Given a foliation on M we define an equivalence relation on M . It is

generated by setting a and b equivalent if they are in the same flow box and
lie on the same local leaf of that flow box. Each equivalence class is a leaf
of the foliation. Each leaf is the image of a smooth one-to-one immersion
of a smooth k-dimensional manifold to M . (The map is not necessarily an
embedding since the image may not be a closed subset.
Notice that a k-dimensional foliation F in M determines a k-dimensional
distribution, the tangent distribution to the leaves of the foliation. The
theorem about integrating vector fields implies that every 1-dimensional
distibution is the tangent distirbution to a foliation. (If the distribution
is orientable, then choose a section producing a non-=where zero vector
field which can then be integrated to give a 1-dimensional foliation. If the
distribution is not orienatble, pass to the double cover where it is orientable.
Then taking a non-where zero section and integrating gives a flow on the
double covering that is tangent to induced distribution on the double cover.
Taking the quotient by the covering transformation produces a foliation on
the quotient as required.)
Consider the 2-dimensional distribution on R3 given by the kernel of the
1-form dz − xdy. At (x, y, x) the plane is spanned by {∂x , ∂y + x∂z }. In
particular, the two-planes in this distribution project isomorphically onto
the (x, y)-plane. This distribution is not tangent to a foliation. If it were
then every leaf of the foliation would project isomorphically onto the (x, y)-
plane. But if we start and move along y-axis to (0, 1, 0) the integral curve
tangent to the distribution is the interval on the y-axis. Starting at (0, 1, 0)
and moving parallel to the x-axis for a unit distance the integral curve ends

27
Figure 1: Foliations

28
at (1, 1, 0). On the other hand, if we first move along an integral curve
projecting to the x-axis a unit length we end at (1, 0, 0). Then the integral
curve projecting to an interval of length 1 parallel to the y-axis ends at
(1, 1, 1). Since these endpoints do not agree, there can be a foliation tangent
to the distribution.
Here is the result that tells us when a distribution can be integrated to
a foliation (i.e., is tangent to a foliation).

Theorem 3.3. (Frobenius Theorem) Let D be a k-dimensional distribution

on a smooth manifold M . The D is the tangent distribution to a foliation
if and only if the space of vector fields whose values are contained in D is
closed under Lie bracket.

Definition 3.4. Distributions that are closed under Lie bracket in the sense
given in the theorem are called involutive.

Proof. First let us show that the tangent distribution to a foliation is invo-
lutive. The result is local so we may as well work in a flow box U k × V n−k ⊂
Rk × Rn−k with the leaves being given by U k × {y} as y varies in V . Then
a vector field tangent to the foliation is of the form
k
X
µi (x, y)∂xi .
i=1

Clearly, the Lie bracket of any two such vector fields is again a vector field
in the space spanned over the smooth functions of the {∂xi }1≤i≤k .
For the converse, I will prove the case when the dimension of the foliation
is 2. This contains the essential idea and the higher dimensional proof
requires only more intricate record keeping. Again we work locally. Given
a 2-dimensional distribution D and a point x ∈ M we can choose local
coordinates U 2 × V n−2 near x that are open balls centered at the origin in
R2 and Rn−2 , respectively, so the projection of the two-planes of D to the
tangents planes to first factor are isomorphisms. We lift ∂1 , ∂2 to vector fields
χ1 , χ2 in D. These generate D throughout this region. Since [χ1 , χ2 ] ∈ D,
it is a linear combination over the smooth functions of χ1 and χ2 and hence
the bracket is zero if and only if its projection to the (x1 , x2 )-plane is zero.
But since the projection of χi is ∂i , the projection of [χ1 , χ2 ] to this plane is
zero. Thus, [χ1 , χ2 ] = 0. That is to say that, locally at least D is generated
by two commuting vector fields.
Now we define coordinates on the subspace {x2 = 0} by integrating χ1 .
That is to say we use the given coordinates (x3 , . . . , xn ) on {x1 = x2 = 0}

29
and then use the integral curves of χ1 to extend these n−2 coordinates to the
subspace {x2 = 0} by requiring these coordinates to be constant on the in-
tegral curves. These extension together with x1 define new local coordinates
on {x2 = 0}. In these coordinates χ1 = ∂1 along the subspace {x2 = 0}.
Now we do the same thing with χ2 extending these n − 1 coordinates to
an entire neighborhood by requiring them to be constant along the integral
curves of χ2 . These, together with the coordinate x2 define a full coordinate
system throughout a small neighborhood of x. In these coordinates χ1 = ∂1
along x2 = 0 and χ2 = ∂2 .
The
P condition on χ1 means that in these coordinates it is written as
i i 2
χ1 + i f (x)∂ P i withi f = 0 along {x = 0}. The condition [χ1 , chi2 ] = 0
now reads − i ∂2 (f )∂i = 0, meaning that ∂2 (f i ) = 0 for all i. Since f i = 0
along {x2 = 0}, it follows that f i = 0 for all i; that is to say χ1 = ∂1 in
these coordinates. Thus, the foliation tangent to the distribution in this
neighborhood is given by the family of surfaces {(x3 , . . . , xn ) = constant},
which is the usual 2-dimensional foliation of R2 × Rn−2 restricted to this
neighborhood.

3B: Integration of Differential Forms

We have seen where the term differential comes from, now let us explain
where the term ’form’ comes from. Originally, the expressions we now call
differential forms were of the correct ‘form’ to be integrated (over oriented
manifolds).

3.1 Integration
3.1.1 The case of smooth curves in a smooth manifold
Consider a differential 1-form ω on a smooth manifold M and a smooth map
γ : [0, 1] → M . Then γ ∗ ω is a differential 1-form on [0, 1], and thus of the
form g(t)dt. This is exactly the type of expression we can integrate: We
define Z Z 1
ω= g(t)dt.
γ([0,1]) 0

The key point here is that this definition is independent of the parameteri-
zation of the domain interval of γ (as long as the parameterization is to have
the same initial and terminal points). For suppose that ϕ : [a, b] → [0, 1] is
an increasing reparameterization (meansing ϕ(a) = 0, ϕ(b) = 1, and ϕ0 > 0

30
everywhere. Then the change of variables formula gives:
Z b Z b Z b
∗ 0
ϕ (g(t)dt) = g(ϕ(s))ϕ (s)ds = (γ ◦ ϕ)∗ ω.
a a a

Notice that if the repazrameterization reverses the direction, then the inte-
gral changes by a multiplicative factor of −1.
The upshot is that differential 1-forms on M can be integrated over
compact, smooth directed curves in M , and the result multiplies by −1 when
we reverse the direction on the smooth curve. The integral of a differential
1-form is additive under juxtaposition of smooth curves.

3.1.2 Orientations of smooth manifolds

A direction on a curve is an example of an orientation of a manifold. Con-
sider a real vector space V of dimension finite dimension n > 0. Then Λn V
is a 1-dimensional real vector space. An orientation for V is the choice of
a non-zero vector in Λn V up to positive multiple; i.e., a choice of the ‘pos-
itive’ direction in Λn V . An ordered basis {e1 , . . . , en } for V determines an
orientation for V given by the element e1 ∧ · · · ∧ en ∈ Λn V . If we switch to
another basis {f1 , . . . , fb } then the effect on the orientation is given by the
sign of the determinant of the matrix expressing the {fi }i in terms of the
{ej }j .
Now consider a vector bundle E → M with fibers n-dimensional real
vector spaces. Then Λn E → M is a line bundle. An orientation for E
is a trivialization of Λn E → M up to a positive function on M ; i.e., an
orientation of each fiber of Λn E that are locally trivial.
An orientation for a smooth manifold M is an orientation for its tangent
bundle. A manifold is orientable if there is an atlas for the smooth structure
so that all overlap functions are orientation preserving maps of between open
subsets of Rn . Of course, as the Möbius band shows not every manifold
is orientable, though of course every manifold is locally orientable in the
sense that every point has an open neighborhood that is orientable. If
the manifold has a boundary, then the orientation of the manifold induces
one on the boundary with the property that a basis for Ty ∂M gives the
orientation of ∂X at y if and only if this basis preceded by an outward
pointing normal gives the local orientation of X at y. This convention
means that the boundary of the unit 2-disk with its orientation induced
from the usual one on R2 is the circle with the orientation given by the
counter-clockwise direction.

31
 A connected zero manifold has two orientations and they are denoted ±.
The value of a 0-form (function) on an oriented 0-manifold is the product
of the value of the function on the point times the sign of the orientation.
With these conventions the boundary of a [0, 1] with its usual orientation is
{1} − {0}. .
Associated to a manifold M is the orientation double covering M f → M.
The points of Mf consist of a point p ∈ M and an orientation of Tp M . The
manifold M is orientable if and only if M
f is isomorphic to the double cover
{±} × M , and a global section of this bundle is equivalent to an orientation
of M .

3.1.3 Integration of compactly supported differential n-forms over

open subsets of Rn
Let U ⊂ Rn be an open subset and let ω be an n-form on U with compact
support. Then
ω = f (x1 , . . . , xn )dx1 ∧ · · · ∧ dxn
for some smooth function f defined on U with compact support. We define
Z Z
ω= f dx1 · · · dxn ,
U U
the usual Riemannian integral of the compactly supported function f on
U . Once again the change of variables formula in integration tells us that
if we have a diffeomorphism ϕ : V → U between open subsets of Rn , with
Cartesion coordinates {y 1 , . . . , y n } on V then
Z Z
∗ 1 n
ϕ (f )|det(Dϕ)|dy · · · dy = f dx1 · · · dxn .
V U
Of course,
ϕ∗ (f dx1 ∧ · · · ∧ dxn ) = (f ◦ ϕ)det(Dϕ)dy 1 ∧ · · · ∧ dy n .
Thus, if the determinant of Dϕ is everywhere positive, then we have
Z Z
∗ 1 n
ϕ (f dx ∧ · · · ∧ dx ) = ϕ∗ (f )det(Dϕ)dy 1 ∧ · · · ∧ dy n
V ZV

= ϕ∗ (f )det(Dϕ)dy 1 · · · dy n
ZV

= f dx1 · · · dxn
ZU
= f dx1 ∧ · · · ∧ dxn .
U

32
 Similarly, if detDϕ is everywhere negative, then
Z Z
∗ 1 n
ϕ (f dx ∧ · · · ∧ dx ) = − f dx1 ∧ · · · ∧ dxn .
V U

In the above discussion we have assumed that we are working with an

open subset U of Rn but the arguments easily adapt to the case when U
is an open subset of the half-space {x1 ≥ 0}, and even to manifolds with
corners given as open subsets of {x1 ≥ 0, . . . , xk ≥ 0}.

3.1.4 Integration over Compact Submanifolds

Suppose that M is a smooth n manifold and ω is a differential k-form on
M . Let X be a compact,oriented k-manifold, possibly with boundary,. We
can cover X by finitely many coordinate charts U1 , . . . , UN of M with the
property that Ui ∩X is the intersection of Ui with a coordinate k-dimensional
subspace given by{xj = 0} for all j > k, or a half-space therein and the
orientation of X agrees with the usual orientation of this k-dimensional
subspace. Extend the {Ui }1≤i≤n to a covering of M by coordinate charts so
that all the other charts are disjoint from X. Choose a partition of unity
{ψi } subordinate to this covering. Then
N
X
ω|X = (ψi ω)|Ui .
i=1

The differential form (ψi ω)|X is then supported in the coordinate chart
Ui ∩ X.R It is a differential k-form on an open subset of a coordinate patch
Hence, Ui ∩X (ψi ω)|X is defined as above
R
Then the integral X ω is defined as the sum of these integrals. It is an
easy exercise using the fact that the individual terms are independent of the
choice of Euclidean coordinates to show that the result is independent of all
choices of open covering and partition of unity.
This gives us an integration pairing between differential k-forms and
compact smooth k-dimensional submanifolds, possibly with boundary (or
even corners), of M .

3.2 Stokes’ Theorem

Theorem 3.5. Let M be a smooth n-manifold and let X ⊂ M be a smoothly
embedded k-dimensional oriented submanfiold with boundary. Then for any

33
differential (k − 1)-form ω on M we have
Z Z
dω = ω,
X ∂X

when we use the induced orientation on ∂X.

Proof. By restricting to X it suffices to assume that M = X. Next, let us

consider the case when X is one dimensional and ω is a smooth function f .
If X is an interval with [a, b] with its induced orientation, Stokes’ Theorem
becomes Z b
f 0 (t)dt = f (b) − f (a),
a
the fundamental theorem of calculus. If X is a circle, then we take a ‘pa-
rameterization’ by the interval [0, 2π]. Then f (2π) = f (0) since 0 and 2π
map to the same point of X. Thus,
Z Z 2π
df = f 0 (t)dt = 0.
X 0
R
In this case ∂X = ∅, so that ∂X f = 0. So once again the result comes
down to the fundamental theorem of calculus.
Now we turn to the higher dimensional case. We cover X by coordinate
patches. {U1 , . . . , UP
N }. Using a partition of unity subordinate to this cover
we can write ω = i ωi where for each i the form ωi has compact support
contained in Ui . In fact, we Q can suppose that for each i the support of ωi
is supported in a box Bi = kj=1 [a, b] in the Cartesian coordinates on Ui .
Furthermore, we can assume that for each i either the box Bi is contained
in the interior of X or it meets ∂X exactly along (n − 1)-face {b} × Bi0 where
Bi0 = kj=2 [a, b]. For boxes Bi in the interior of X, the differential form ω
Q
vanishes on ∂Bi . For boxes Bi that meet ∂X the differential form vanishes
except possibly on the P face {b} × Bi0 .
We write ωi = I Ij means the wedge product, in
j gi,j dx , where dx
j

order, of the dx1 , . . . , dxk omitting dxj . Let us consider the germ gi,1 dxI1 .
Its differential is ∂1 gi,1 dx1 ∧ · · · ∧ dxk and by Fubini’s theorem we have
Z Z Z b 
d(gi,1 dxI1 ) = (∂1 gi,1 )dx1 dx2 dx3 · · · dxk .
Bi Bi0 a

34
By the fundamental theorem of calculus the right-hand side is equal to
Z  
gi,1 (b, x2 , . . . , xk ) − gi,1 (a, x2 , . . . , ak ) dx2 · · · dxk =
Bi0
Z
= gi,1 (b, x2 , . . . , xk )dx2 · · · dxk
Bi0
Z
= gi,1 dxI1 ,
∂X

where the first equation comes from the fact that g(a, x2 , . . . , xk ) is identi-
cally zero.
I`
R
A similar argument shows that for the cases ` > 1 both ∂X gi,` dx and
I
R
X d(gi,` dx ) are zero.
`

Definition 3.6. A differential form ω is said to be closed if dω = 0 and is

said to be exact if there is a differential form η with dη = ω.
Notice that since d2 = 0 every exact form is closed.
Corollary 3.7. 1. Suppose that ω is a closed form and X ⊂ M is a closed
submanifold which is the boundary of an oriented submanfiold Y ⊂ M . Then
Z
ω = 0.
X

2. Suppose that ω is an exact form and X is a closed oriented submanifold

then Z
ω = 0.
X
Proof. For 1 Z Z Z Z
ω= ω= dω = 0 = 0.
X ∂Y Y Y
For 2, write ω = dη. Then
Z Z Z
ω= dη = η = 0,
X X ∂X

where the last equality is a consequence of the fact that ∂X = ∅.

Corollary 3.8. Let ω be a closed differential k-form and X a closed, ori-

ented smooth manifold. Suppose that i1 : X → M and i2 : X → M are
smooth embeddings that are homotopic. Then
Z Z
ω= ω.
i1 (X) i2 (X)

35
Proof. We can approximate the homotopy between i1 and i2 by a smooth
homotopy between these embeddings. That is to say there is a smooth map
J : X × I → M with J|{j}×X = ij for j = 0, 1 Then J ∗ ω is a closed form on
I × X. Hence, Z Z Z
0= dω = ω− ω.
I×X i1 (X) i2 (X)

3.3 de Rham’s Theorem: A brief introduction

Let Ω∗ (M ) denote the differential graded algebra of differential forms on a
smooth manifold M . Recall that a form ω is closed if dω = 0 and it is exact
if there is η with dη = ω. The fact that d2 = 0 implies that every exact form
is closed. Clearly, the closed forms (being the kernel of a linear map) and
the exact forms (being the image of a linear map) are real vector spaces,
with the second bin a subspace of the first. the em k th -de Rham cohomology
of M is defined as the quotient of closed k-forms modulo exact k-forms

k Ker(d : Ωk (M ) → Ωk+1 (M ))
HdR (M ) = .
Im d : Ωk−1 (M ) → Ωk (M )

[Since the kernel and image are infinite dimensional vector spaces with
(various) natural topologies one might be concerned about for example
whether the image of d is is a closed subspace. But these turn out not
to be real issues.]
de Rham’s Theorem compares this cohomology with algebraic topology’s
singular cohomology. For any toological space one defines the singular ho-
mology by forming a chain complex of abelian groups ⊕k Singk (X). The
group Singk (X) is the free abelian group generated by the continuous maps
of the k-simplex to X. The boundary map ∂ : Singk (X) → Singk−1 (X) is
determined setting the boundary of ϕ : ∆k → X equal to the linear combi-
nation
Xk
(−1)i ϕ|ith −face of ∆k .
i=0

This defines a chain complex (i.e., ∂ 2 = 0). The singular homology of X is

the homology of this chain complex: namely

Ker∂ : Singk (X) → Singk−1 (X)

Hk (X) = .
Im ∂ : Singk+1 (X) → Singk (X)

36
 For any abelian group A one defines the singular chomology of X with co-
efficients in A by taking the cochain complex whose groups are Singk (X; A) =
Hom(Singk (X), A) with differential
d : Singk (X; A) → Singk+1 (X; A)
to be the dual to the boundary map in the singular chain complex. Then
d2 = 0 and the singular cohomology
Ker d : Singk (X; A) → Singk+1 (X; A)
H k (X; A) = .
Im d : Sing−1 (X; A) → Singk (X; A)
The basic idea for comparing de Rham cohomology with singular coho-
mology comes from the pairing of integrating differential forms over chains.
There is a technical issue in that the objects in singular cohomology have no
smooth requirements (since they are defined for arbitrary topological spaces
not just smooth manifolds.
There are many ways to deal with this problem. One is to consider
smooth chain complex of a smooth manifold. Here the chain groups are
the free abelian groups generated by smooth maps of the simplicies into
a smooth manifold M . The same boundary formula works in this case to
show that we have an analogous smooth singular chain complex of a smooth
manifold, which we denote by SSing∗ (M ). Now the integration pairing is
defined:
Ωk (M ) ⊗ SSingk (M ) → R
R
given by ω ⊗ A 7→ A ω. We can view this as an R-linear map

Ωk (M ) → SSingk (M ; R).
Stokes’ Theorem is the statement that this map of cochain complexes is a
cochain map (i.e., it computes with the differentials). As a result it defines
a linear map
k k
HdR (M → HSSing (M ; R)
where the right-hand term is the cohomology of the smooth singular cochain
complex with R-coeffiients.
de Rham’s Theorem consists of two isomorphisms;
Theorem 3.9. 1. The integration pairing above determines an isomorphism
k ∼
= k
HdR (M ) −→ HSSing (M ; R)
2. The inclusion of the chain complex SSing∗ (X) → Sing∗ (X) induces an
isomorphism on homology.

37
 I shall not give a proof of de Rham’s Theorem in this course.
It follows from the second result that the inclusion map SSing∗ (M ) →
Sing∗ (M ) induces isomorphisms on the associated cohomology with any co-
efficients, in particu;lar with R-coefficients.

Corollary 3.10. The de Rham cohomology of a smooth manifold is identi-

fied with its singular cohomology with R-coefficients. This identification is
natural for smooth maps between smooth manifolds.

Even more is true: Wedge product of diferential forms induces a multipli-

cation de Rham cohomology making it a graded commutative algebra over
R. There is for singular cohomology a cup product formula due to Whitney.
This makes singular cohomology with integer coefficients a graded commuta-
tive ring and hence makes singular cohomology with R-coeficients a graded
commutative algebra over R. The isomorphism of de Rhan’s theorem is a
ring isomorphism.

38
4 Complex plane curves and holomorphic one-forms
4.1 The Complex Curves
In this lecture we shall study curves in C2 given by a single complex poly-
nomial equation C = {P (x, y) = 0}. We will only consider those curves
that are smooth, i.e., curves given by polynomials P (x, y) with the property
that dP 6= 0 at every point of C. The equations we shall study are of the
form P (x, y) = y 2 − p(x). Equivalently, we are considering curves given by
the equation y 2 = p(x). In this case the condition that the curve C defined
by this equation is smooth is that p(x) has only simple roots, meaning that
there is no a ∈ C such that p(x) factors as (x − a)2 q(x). We make this
assumption from now on in this lecture.
The projection C2 → C sending (x, y) to x induces a holomorphic map
π : C → C that is 2-to-1 everywhere except at the roots of p where there is
only one point in the pre-image. Furthermore, for any (x, y) ∈ C with y 6= 0
(or equivalently x not a root of p), the differential of π is an isomorphism so
that x is a local coodrinate locally at any point of C \ C ∩ {y = 0}. At the
remaining points y is a local coordinate. The reason is that ∂x (y 2 − p(x)) =
−p0 (x) does not vanish at (0, x) ∈ C since the roots of p(x) are simple.
Near (0, x) ∈ C the projection to C → {y = 0} looks like z 7→ z 2 + a
in appropriate local coordinates on C. The point is that we can write y 2 =
(x−a)q(x) where q(a)p 6= 0. Thus, locally we can take an holomorphic square
root of q(x), denoted q(x) and
!2
y
p = (x − a).
q(x)
By the implicit function theorem we know that q(x) is a non-zero holo-
mor[phic function of y near (0, a), hence so is its square root. Hence we can
use
y
z=p
q(x)
as a local holomorphic coordinate on C near (0, a) and (x − a) = z 2 so that
π(z) = z 2 + a. In particular the pre-image in C of a small disk centered
at a is a disk centered at the origin in the y1 coordinate on C. These
points are branch points for the map to x-axis. Centered at any other point
a ∈ C distinct from the roots of p(x) = 0 there is a small disk in the x-
axis whose preimage under π is two disjoint disks in C, each mapping down
by a holomorphic isomorphism. (This is the statement that x is a local
coordinate at all such points of C that do not project to a root of p(x).)

39
Definition 4.1. The p algebraic curve C is called the Riemann surface of the
holomorphic function p(x). Indeed, C is displayed over the x-axis and has
on it a holomorphic function y. Then p values of y on the points above a in
the x-axis all the possible values of p(a). Usually there are two of them,
but for a a root of p there is only one, y = 0.

4.1.1 Nature of the curve near infinity

Lemma 4.2. Let γ be a smooth simple closed curve in the x-axis that does
not pass through any root of p. The pre-image γe = π −1 (γ) maps to γ two-to-
one by a local diffeomorphism. There are two possibilities: γe is itself a loop
and under π it wraps twice around γ or γ e is a disjoint union of two loops
each of maps diffeomorphically onto γ. If the disk in the x-axis bounded by
γ contains an odd number of points the first case holds, otherwise the second
case holds.

Proof. The simple closed curve is homotopic in the complement of the roots
of p(x) to the simple closed curve as drawn in the upper part of Figure
1. Since homotopic loops are in the same case, it suffices to consider the
smaller loops near the arcs connecting pairs of roots as in this figure. Since
traversing the boundary of a disk centered at a root interchanges the sheets
of of C above the x-axis, the result follows easily.

Corollary 4.3. If p is of even degree, then the pre-image of any sufficiently

large loop in the x-axis is two loops in C. If p is of odd degree, then the
pre-image of any sufficiently large circle in the x-axis is a single loop.

Proof. The disk bounded by a sufficiently large loop contains all the roots
of p and the number of roots of p is its degree.

Corollary 4.4. Let γ be a sufficiently large circle in the x-axis. Then the
complement of the disk bounded by γ is an annulus A and its pre-image in
C is either two annuli when p has even degree or a single annulus double
covering A. Thus, a neighborhood of infinity in C is either two punctured
disks each mapping isomorphically onto A when p has even degree, or a single
punctured disk mapping by z 7→ z 2 onto A (in appropriate local coordinates).

In the case when the degree of p is even we compactify the curve C by

adding the centers of the two punctured disks. In the case when the degree
of p is odd we compactify the curve C by adding the center of the punctured
disk neighborhood of infinity. The resulting curves maps two-to-one to CP 1

40
Figure 2: Simple Closed curves in the x-axis

41
which is a local diffeomorphism at each point in the preimage of ∞ when
the degree of p(x) is even and is locally of the form z 7→ z 2 when the degree
of p is odd. In the latter case ∞ ∈ CP 1 is a branch point.

Claim 4.5. When p has even degree, the coordinate ζ = x−1 is a√local
coordinate near each of the completion points. When p has odd degree x−1
is a local coordinate at the completion point. Thus, the compactified curves
are complex curves with a generically two-to -one holomorphic map to CP 1 ,
the completion of the x-axis. If the degree of p is 2d or 2d − 1 this projection
has 2d branch points.

Proof. Since each point at infinity has a neighborhood in C that maps by

holomoprhic isomorphism onto the complement of a large disk centered at
the origin
√ in the plane, the first statement is clear. In the second case
ψ = ζ is a local coordinate on a neighborhood of infinity in C and the
missing point is the origin of the ψ disk.

From now on C refers to the completed curve as defined above.

4.1.2 The Topology of the Curve C

We are considering the curve C = {y 2 = p(x)} in C2 and its compactification
over the compactification CP 1 of the x-axis. First we consider the case when
p(x) has even degree, say 2d. Pair up the roots on p(x) and draw disjoint
arcs between each pair. We get a picture as in the lower half of Figure 1. The
pre-image of a small neighborhood of an arc connecting two of the roots of p
is an annulus and the pre-image of the complement of these d neighborhoods
is two copies of a 2-sphere with d small disks removed. Each of the d annuli
connects the two differ copies. The result is pictured in Figure 2. It is a
surface with (d − 1) holes, called the surface of genus d − 1.
In the case when p is of odd degree 2d − 1, the point ∞ in CP 1 is also
a branch point but the analysis above still holds. Pair up the branch points
and draw disjoint arcs between them on CP 1 . As before the preimages of
neighborhoods of the arcs are annuli, and the pre-image of the rest of Cp1
is two copies of S 2 with d disks removed. Thus, once again the surface is a
surface of genus d − 1.

4.2 Homolomorphic differential 1-forms

Given a smooth curve C ⊂ C2 , with linear coordinates (x, y) on C2 , given
by an equation y 2 = p(x) (with no repeated roots for p), we define a holo-

42
Figure 3: Surface when d = 3

43
morphic differential
dx
.
y
This expression makes good sense near any (a, b) ∈ C with b 6= 0, for at
such points x is a local coordinate and y 6= 0. We have to examine what
happen near points of the form (a, 0) ∈ C. Of course, on the open subset of
C where both x and y are local coordinates we have the equation

2ydy = p0 (x)dx.

Since p0 (a) 6= 0 (this is the no repeated root condition), in a small neighbor-

hood of (a, 0) ∈ C, y is a local coordinate and x is a local coordinate except
that (a, 0). Thus, in the punctured neighborhood of (a, 0) we have

dx 2dy
= 0 .
y p (x)

On the other hand, since p0 (a) 6= 0, the expression

dy
2p0 (x)

is holomorphic through this neighborhood. This shows that dx/y has a

unique holomorphic extension to all of C. By a slight abuse of notation we
denote this global form on C by dx/y.
Let us see what happens at infinity .
Case (i): The degree of p is even, say 2d. We introduce the local
coordinate ζ = x−1 which is a local coordinate near every point of C mapping
to a small neighborhood of ∞ in CP 1 . Then differential one-form dx/y near
infinity is −ξζ −2 dζ where ξ = y −1 .
The function y 2 = ζ −2d r(ζ) where r is a polynomial with r(0) 6= 0, and
hence y = ζ −d h(ζ) where h(ζ) is holomorphic with h(0) 6= 0. There are
two choice for the square root and the two functions correspond to the two
sheets of C over this disk in ζ. Each of the resulting functions ξ then is
holomorphic in this region with a zero of order d = deg(p)/2 at {ζ = 0}.
This implies that (dx/y) = −ξζ −2 dζ has a zero of order d − 2 at infinity.
In each sheet of the pre-image of a small neighborhood ∞ ∈ CP 1 , the
differential one-form dx/y has a zero of order d − 2. This means that when
the degree of p is 2, the form has a simple pole at both pre-images of ∞ and
is meromorphic rather than holomorphic. When the degree of p is four, dx/y
is everywhere holomorphic and non-zero. When the degree of p is 2d > 4,

44
the form dx/y is everywhere holomorphic, has a zero of order d − 2 at each
of the pre-images of ∞ ∈ CP 1 and is otherwise non-zero.
For 2d > 4 we can create similar holomorphic differential one-forms by
taking
q(x)dx
y
where q is a polynomial of degree ≤ d − 2. This produces a vector space of
dimension d − 1 of global holomorphic differentials. Direct computation (see
the Problem Set) shows that this is all the global holomorphic differential
one-forms. on C.
Case (ii): The degree of p is odd, say 2d−1. Now a local coordinate near
the unique point at infinity is z projecting to ζ 2 = x−2 . In this coordinate
the form dx is −2z −3 dz and p(x) = ζ −(2d−1) r(ζ) = z −(4d−2) r(z 2 ), where r
is a polynomial with r(0) 6= 0. Thus,

y = z −(2d−1) h(z)

where h is a holomorphic function with h(0) 6= 0. It follows that in this local

coordinate
dx −2z −3 dz
= −(2d−1) = −2z 2d−4 h−1 (z)dz.
y z h(z)
dx
This means that y has a zero of order 2d − 4 at infinity. Hence the form

q(x)dx
y

with q a polynomial of degree at most d − 2 is holomorphic on all of C. (If

q(x) is a polynomial of degree k, then it creates a meromorphic function on
the z-disk with a pole of order 2k.)
Thus, in either case degree of p is 2d or 2d − 1 all differential forms of
the type q(x)dx
y with q(x) a polynomial of degree at most d − 2 are global
holomorphic differentials on the compact curve C. This gives is a complex
vector space of dimension d − 1 of holomorphic 1-forms.

Claim 4.6. Holomorphic 1-forms on C are closed. No non-zero holomor-

phic 1-form on C can be exact.

Proof. Let ω be a non-zero holomorphic differential form. Since ω is locally

of the form ψ(z)dz with ψ homomorphic, it is clear that ∂z ψ(z) = 0 and
hence dz ∧ ∂z ω = 0. Since the complex dimension of C is one, it follows that
dz ∧ ∂z ω = 0. Hence, dω = 0.

45
 Proof 1. If f is a complex-valued smooth function with df = ω, where ω
is a holomorphic 1-form, then in local holomorphic coordinate on C, we have
∂z f = 0, meaning that f is a holomorphic function. Since C is compact, f
is contant and hence df = 0.
Proof 2. The differential form ω is a closed diffierential form and ω ∧ ω is
a two-form which in local holomorphic coordinates is of the form |f |2 dz ∧ dz
and dz ∧ dz = (−2i)dx ∧ dy. This means that iω ∧ ω is a two-form and
Z
iω ∧ ω > 0.
C

If ω were exact then by the Leibnitz rule iω ∧ ω is also exact and would have
integral 0 over C.

Since the wedge product of two holomorphic 1-forms is zero, it follows

that there is no sum ω1 +ω2 is exact, in particular the direct sum of the space
of holomorphic differentials and the space of their conjugates (called anti-
holomorphic differentials) is a space of closed 1-forms that injects into the de
Rham cohomology HdR 1 (C; C). The image of the holomorphic differentials

is denoted
H 1,0 (C) ⊂ HdR1
(C; C)
and the image of the space of anti-holomorphic differentials is denoted

H 0,1 (C) ⊂ HdR

1
(C; C).

We have just shown that when p has degree 2d or 2d − 1 the dimension of

H 1,0 and H 0,1 are d − 1 and that there is a linear embedding

H 1,0 (C) ⊕ H 0,1 (C) ⊂ HdR

1
(C; C).

On the other hand, when p has 2d or 2d − 1, C is a (d − 1)-holed surface.

Thus, its singular homology of C is a free abelian group of rank 2(d − 1).
We have also seen that integrating over smooth simple closed curves gives
an injection
1
HdR (C; C) → Hom(H1 (C), C).
Since H 1,0 (C) ⊕ H 0,1 (C) and Hom(H1 (C), C) are complex vector spaces of
dimension 2d − 2 and the map between them is an injection, it follows that
∼
= ∼
=
H 1,0 (C) ⊕ H 0,1 (C) −→ HdR
1
(C; C) −→ Hom(H1 (C), C).

46
 In particular, every closed 1-form on C with complex coefficients differs
by an exact form from the sum of a holomorphic and anti-holomorphic 1-
form. Since complex conjugation on HdR 1 (C; C) interchanges H 1,0 (C) and

H 0,1 (C), it follows that H 1,0 (C) and H 0,1 (C) meet HdR
1 (M : R) only at {0}.

There is a real linear isomorphism H1 (M ; R) → HomHdR 1 (C; R), R)

and Hom(HdR 1 (C; R), R) embeds in Hom(H 1 (C; C), C) as those homomor-
dR
phisms ϕ that satisfy ϕ(ω) = ϕ(ω). Since H 0,1 (C) = H 1,0 (C), it follows
that any non-zero ϕ ∈ Hom(HdR 1 (C; R), R) is non-trivial when restricted to
1,0
H (C). Counting dimensions, we see that the map
1
Hom(HdR (C; R), R) → Hom(H 1,0 (C), C)

is a real linear isomorphism. This means that

Hom(H 1 (C; Z), Z) → Hom(H 1,0 (C), C)

is an embedding onto a lattice of full rank (2(d−1)) in Hom(H 1,0 (C), C). Of
course, Hom(H 1 (C; Z), Z) = H1 (C). Thus, we have an embedding H1 (C; Z)
as a lattice of full rank in Hom(H 1,0 (C), C). Tracing through this map it is
given as follows: the homology class of curve γ maps to the homomorphism
that sends a holomorphic differential ω to
Z
ω.
γ

Choosing an integral basis γ1 , . . . , γ2(d−1) for H1 (C) and a complex basis

ω1 , . . . , ωd−1 for H 1,0 (C) we have a (d − 1) × 2(d − 1) matrix
Z
ωj .
γi

The 2(d − 1) columns of this matrix generate a lattice of rank 2(d − 1) in

the Cd−1 ; i.e., these 2(d − 1) vectors are linearly independent over R.
The quotient of Cd−1 by this lattice is a complex torus, called the Jaco-
bian of the curve and denoted J(C) call the Abel-Jacobi map. Integarting
the holomorphic differentials along paths from the base point x0 ∈ C defines
a holomorphic map C → J(C). The Jacobian J(C) and the Abel-Jacobi
map C → J(C) exist for all compact smooth complex curves and are an
essential ingredient in the study of complex structures on Riemann surfaces.

47